FIGS. 1 and 2 are diagrams illustrating the Hough transform, which is conventionally used for extracting information on a straight line, from information on an image. According to the Hough transform, when a set of coordinates of a pixel the value of which corresponds to black (or white) of an original image is denoted by (xi, yi), curves expressed by the equations EQU .rho.=xisin.theta.+yicos.theta.,
are drawn on a .rho.-.theta. plane (FIG. 2) for all of the pixels the values of which correspond to black (or white). Since points on the same straight line on the original image (x-y plane) correspond to the same set of coordinates .rho., .theta. as indicated in FIG. 1, the above curves cross at a crossing point (.rho., .theta.) on the .rho.-.theta. plane (FIG. 2). Therefore, the straight line on the original image (x-y plane) can be obtained from the above crossing point of the curves on the .rho.-.theta. plane.
However, it is necessary to draw the curves as above for all of the points on each straight line, to obtain a straight line in accordance with the above method, and processing to obtain a crossing point is necessary for each straight line. Therefore, the amount of data processing becomes great. This great amount of data processing takes a large amount of software processing time, and large size hardware for the hardware processing.
Further, there is a drawback that only straight lines can be detected by the Hough transform, and it is impossible to detect a curve which can be locally deemed to be a line segment, but macroscopically has a curvature.